Context-guided multi-turn interviews reveal more VLM hallucinations than static benchmarks, and those hallucinations increase with conversational history and false-premise questions.
V ogelstein, Brian Gallagher, and Christos Faloutsos
1 Pith paper cite this work, alongside 146 external citations. Polarity classification is still indexing.
1
Pith paper citing it
146
external citations · OpenAlex
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Contextualized Evaluation of Vision Language Models through Dynamic, Multi-turn Interactions
Context-guided multi-turn interviews reveal more VLM hallucinations than static benchmarks, and those hallucinations increase with conversational history and false-premise questions.